I built AI Dimag after hitting this wall repeatedly: I'd spend 30 minutes explaining architectural decisions to Claude or Cursor, then start fresh in new chat the next day and have to explain everything again.
Common things that get forgotten:
"All database access goes through src/db/store.ts"
"We tried microservices approach X and it failed because Y"
"Never use library Z, it has security issues"
"Always run tests before committing to main"
The frustrating part? These aren't just preferences they're critical decisions with context that took hours to figure out.
AI Dimag gives MCP-compatible agents (Claude, Cursor, Copilot, Windsurf) persistent memory that survives across sessions. Unlike traditional context systems, every memory carries falsifiable evidence—shell checks, tests, commit anchors—that dim verify re-runs against your current repo. Stale beliefs get flagged instead of silently misleading your AI. Works via MCP tools for real-time memory or generates static context files (.cursorrules, CLAUDE.md) for non-MCP tools. Free for teams ≤10.
